Work progressed in two areas: (1) development and characterization of a new method of analysis for statistically dependent outcomes; and (2) a new design for case-control studies, based on pooling of specimens for assay. Statistically-dependent data arise in many research contexts. In toxicology, pup-specific outcomes may be more similar for pups from the same litter than for pups from different litters, causing the "litter effects" problem. When assessing effects of exposures on risk of miscarriage, pregnancy outcomes tend to be more similar among pregnancies from a given woman, because of variability in baseline risk or response from woman to woman. We have developed a method based on sampling one outcome per cluster, carrying out a classical analysis on these now-independent outcomes, and then repeating the resampling many times, ultimately pooling the parameter estimates. "Within Cluster Resampling" (WCR) performs well in simulations, is analytically simple, and avoids the need to make ad hoc assumptions about the underlying covariance structure. A separate project is related to the problem of designing case-control studies in a context where biologic specimens are available, but the assay is expensive. The "Pooled Exposure Assessment" design (PEA) pools equal aliquots from randomly grouped (disjoint) sets of cases and sets of controls. Under a multiplicative model for risk, these "pooling sets" can be analyzed to produce valid estimates of the person-based dose-response for risk. We showed that very little statistical power is sacrificed. Savings are potentially great for a study involving an expensive assay, such as that required for measuring dioxin in serum.